Researchers at UCL’s Institute for Neurology have leveraged AI language models to detect subtle speech patterns in patients with schizophrenia. Published in PNAS, their study explores the potential of AI-driven language analysis for psychiatric diagnosis and assessment.
Psychiatric diagnoses traditionally rely heavily on verbal interactions, with minimal reliance on objective tests such as blood work or brain scans. This limitation hinders our understanding of the causes of mental illness and the monitoring of treatment progress.
In this study, 26 participants with schizophrenia and 26 without the condition completed verbal fluency exercises. Using an AI language model trained on a vast corpus of internet text, the researchers assessed the predictability of words recalled by participants. The control group’s responses were notably more predictable than those of schizophrenia patients, especially in cases with severe symptoms.
The researchers propose that this divergence may be linked to the brain’s formation of associations between memories and ideas, a theory supported by brain scanning data from the same study.
Dr. Matthew Nour, the lead author, highlights the transformative potential of AI language models in the field of psychiatry, which is inherently intertwined with language and meaning.
Schizophrenia, a widespread psychiatric disorder, affects millions worldwide. Common symptoms include hallucinations, delusions, cognitive disarray, and behavioral changes.
The UCL and Oxford research team plans to expand their study with a broader patient sample and in various speech settings to explore clinical applications of this technology. Dr. Nour anticipates the integration of AI language models into medical practice in the coming decade, offering a promising future for neuroscience and mental health research.